Frankel, Joe

Magimai-Doss, Mathew

King, Simon

Livescu, Karen

Çetin, Ozgur

2007-09-18T10:02:10Z

2007-09-18T10:02:10Z

2007

The so-called tandem approach, where the posteriors of a multilayer perceptron (MLP) classifier are used as features in an automatic speech recognition (ASR) system has proven to be a very effective method. Most tandem approaches up to date have relied on MLPs trained for phone classification, and appended the posterior features to some standard feature hidden Markov model (HMM). In this paper, we develop an alternative tandem approach based on MLPs trained for articulatory feature (AF) classification. We also develop a factored observation model for characterizing the posterior and standard features at the HMM outputs, allowing for separate hidden mixture and state-tying structures for each factor. In experiments on a subset of Switchboard, we show that the AFbased tandem approach is as effective as the phone-based approach, and that the factored observation model significantly outperforms the simple feature concatenation approach while using fewer parameters.

100851 bytes

application/pdf

J. Frankel, M. Magimai-Doss, S. King, K. Livescu, and O. Çetin. Articulatory feature classifiers trained on 2000 hours of telephone speech. In Proc. Interspeech, Antwerp, Belgium, August 2007.

http://hdl.handle.net/1842/2000

en

speech technology

Articulatory feature classifiers trained on 2000 hours of telephone speech

Conference Paper